TMX: Exchanging Translation Memories

The problem TMX addresses

A translation memory stores reusable relationships between source and target content. A team changing tools may want to keep those relationships rather than begin again with completed documents alone.

The practical question is whether an export preserves the information the receiving workflow needs. Successfully opening a file is only the first check. The imported content also needs to retain its meaning and usable context.

An illustrative handover test

Imagine an English-to-German memory containing approved interface translations. Before a full migration, our recommendation is to create a small representative export containing ordinary sentences, punctuation, inline formatting and relevant metadata.

Import that sample into the receiving tool. Compare the source-target pairings, language identifiers and any fields the team relies on. Where the export does not carry a property, decide whether to transfer it separately or accept its loss explicitly.

This is an original migration exercise, not an official TMX conformance test. It is designed to reveal assumptions before the team moves a large resource.

What TMX does not establish

An exchange format does not prove that the translations are correct, current or appropriate for reuse. A memory can contain obsolete instructions, inconsistent terminology or content approved for a different audience.

Our recommendation is to distinguish technical validation from linguistic validation. An XML file can be structurally acceptable while the stored translation is unsuitable.

LISA and AI

TMX belongs directly to the LISA technical legacy. For AI-assisted work, it can be useful to treat a memory as a curated source of examples rather than as automatically trustworthy material. Preserve provenance and usage permissions, and screen the resource before submitting it to an external system.